Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

The Self-Learning Agent with a Progressive Neural Network Integrated Transformer

Ajay Sivakumar, Shalini, Vasantha Raj, Sebastian Sylvester

TL;DR

The paper tackles catastrophic forgetting in continual learning by integrating Progressive Neural Networks with a pre-trained LLaMA 3.2 transformer and an autonomous data-collection agent to enable task-incremental learning for conversational AI and code generation. It combines Meta-Learning for rapid adaptation, LoRA for efficient fine-tuning, and Elastic Weight Consolidation to preserve prior knowledge, orchestrated by an agent that gathers Wikipedia-derived data. Empirical results show lower perplexities (e.g., 22.1 for conversation and 19.8 for coding) and strong retention across tasks, with BLEU and code accuracy supporting practical performance gains. The approach demonstrates scalable, low-data continual learning and advances toward AGI-like adaptability in dynamic, multi-domain environments, with potential applications in scalable, autonomous AI systems.

Abstract

This paper introduces a self-learning agent that integrates LLaMA 3.2 with a Progressive Neural Network (PNN) for continual learning in conversational AI and code generation. The framework dynamically collects data, fine-tunes tasks with minimal samples, and leverages Meta-Learning for rapid adaptation. LoRA optimizes fine-tuning, while Elastic Weight Consolidation (EWC) enhances knowledge retention. Experimental results demonstrate improved adaptability and memory stability, positioning this approach as a scalable step toward Artificial General Intelligence (AGI).

The Self-Learning Agent with a Progressive Neural Network Integrated Transformer

TL;DR

The paper tackles catastrophic forgetting in continual learning by integrating Progressive Neural Networks with a pre-trained LLaMA 3.2 transformer and an autonomous data-collection agent to enable task-incremental learning for conversational AI and code generation. It combines Meta-Learning for rapid adaptation, LoRA for efficient fine-tuning, and Elastic Weight Consolidation to preserve prior knowledge, orchestrated by an agent that gathers Wikipedia-derived data. Empirical results show lower perplexities (e.g., 22.1 for conversation and 19.8 for coding) and strong retention across tasks, with BLEU and code accuracy supporting practical performance gains. The approach demonstrates scalable, low-data continual learning and advances toward AGI-like adaptability in dynamic, multi-domain environments, with potential applications in scalable, autonomous AI systems.

Abstract

This paper introduces a self-learning agent that integrates LLaMA 3.2 with a Progressive Neural Network (PNN) for continual learning in conversational AI and code generation. The framework dynamically collects data, fine-tunes tasks with minimal samples, and leverages Meta-Learning for rapid adaptation. LoRA optimizes fine-tuning, while Elastic Weight Consolidation (EWC) enhances knowledge retention. Experimental results demonstrate improved adaptability and memory stability, positioning this approach as a scalable step toward Artificial General Intelligence (AGI).

Paper Structure

This paper contains 19 sections, 5 equations, 2 figures, 1 table.

Table of Contents

  1. Introduction
  2. Existing Work
  3. Proposed System
  4. Progressive neural network:
  5. Meta-Learning:
  6. Low-Rank Adaptation (LoRA):
  7. Elastic Weight Consolidation (EWC):
  8. Methodology
  9. Data Collection:
  10. Base Model Training:
  11. Model Architecture:
  12. Fine-Tuning Agent:
  13. Result Obtained
  14. Experimental Setup
  15. Performance Metrics
  16. ...and 4 more sections

Figures (2)

  • Figure 1: Model Architecture.
  • Figure 2: Agent Architecture.